<p>Cervical cancer remains a pressing global health challenge, where timely detection is critical to improving survival outcomes. Machine learning (ML) has emerged as a promising tool for early diagnosis; however, its effectiveness is often constrained by two major challenges in medical datasets: severe class imbalance and class overlap. While imbalance biases classifiers toward the majority class, overlap obscures minority (cancer-positive) cases within ambiguous decision boundaries, leading to high misclassification rates. To address these issues, we propose a hybrid framework— Outlier + Mutual Natural Neighbor Cleaning + Balanced Random Forest (Outlier + MNNC + BRF)—that integrates outlier removal, overlap-aware cleaning, and ensemble learning. First, Isolation Forest is employed to eliminate noisy outliers that distort decision boundaries. Next, Mutual Natural Neighbor Cleaning (MNNC), a density-adaptive under-sampling method, is applied to selectively remove majority instances in overlapping regions, thereby enhancing minority class visibility. Finally, Balanced Random Forest (BRF) is used to mitigate residual imbalance while maintaining robust generalization. The framework is evaluated on a cervical cancer dataset using G-Mean and Balanced Accuracy, with comparisons against ten state-of-the-art resampling methods combined with Random Forest classifiers. Experimental results demonstrate that the proposed method achieves superior performance, attaining a G-Mean of 0.90697 and Balanced Accuracy of 0.908359, outperforming all benchmark techniques. These findings underscore the importance of jointly addressing imbalance and overlap in medical datasets, offering a reliable pathway toward improved early cancer detection.</p>